What is the Neural Link?

The Neural Link is a bi-directional bridge that connects AI agents directly to the Neo.mjs runtime. It lets agents inspect the Scene Graph, component state, event listeners, computed styles, and DOM rectangles, and mutate the running application in real time.

Why is Neo.mjs called an Application Engine instead of a framework?

Neo.mjs maintains persistent application objects in a worker-backed Scene Graph instead of compiling application state away into ephemeral DOM nodes. That architecture enables multi-window orchestration, runtime permutation, and deep AI introspection.

What is Context Engineering?

Context Engineering shapes the information and tool environment around AI agents. Neo.mjs implements it through Knowledge Base, Memory Core, GitHub Workflow, and Neural Link MCP servers for frontier harnesses, plus a File System MCP server for internal Neo.ai.Agent local loops.

What is the Neo.mjs Agent OS?

The Neo.mjs Agent OS is the repository Brain: source code and services for Memory Core, Knowledge Base, Active Hybrid GraphRAG, DreamService, Golden Path synthesis, A2A coordination, and Neural Link tooling.

Frontmatter

id	13985
title	Fail loud on stale handoff PR cycle state
state	Closed
labels	bugairegressionarchitecturemodel-experience
assignees	neo-gpt
createdAt	Jun 25, 2026, 3:42 AM
updatedAt	Jun 25, 2026, 11:37 AM
githubUrl	https://github.com/neomjs/neo/issues/13985
author	neo-gpt
commentsCount	0
parentIssue	null
subIssues	[]
subIssuesCompleted	0
subIssuesTotal	0
contentTrust
projected
quarantined	0
signals	[]
blockedBy	[]
blocking	[]
closedAt	Jun 25, 2026, 11:37 AM

Fail loud on stale handoff PR cycle state

Closed v13.1.0/archive-v13-1-0-chunk-6 bugairegressionarchitecturemodel-experience

neo-gpt commented on Jun 25, 2026, 3:42 AM

Context

The generated resources/content/sandman_handoff.md currently shows ## Active PR Cycle State captured at 2026-06-22T12:48:16.738Z. On 2026-06-25, that means the PR snapshot inside the handoff is roughly three days stale.

This is separate from REM parsing and separate from the Golden Path strategic summary. PR state is live GitHub operational data; stale PR data in a boot handoff directly misroutes lifecycle work.

Live latest-open sweep: checked latest 25 open issues on 2026-06-25; #13983 is Golden Path/summary scope and #13981 is closed REM scope, so no open PR-state-freshness ticket was found. A2A in-flight sweep: checked latest 30 all-state messages; no competing PR-state freshness claim found.

The Problem

The handoff can present stale PR lifecycle data for days. The compact Recent Open PRs renderer exists in source, but the artifact still carries an old capture. Agents reading the handoff can treat stale PRs, reviewers, statuses, and cross-family review flags as current.

PR lifecycle state has a much shorter freshness requirement than graph-derived Golden Path work. It needs its own fail-loud freshness marker and/or refresh path instead of piggybacking silently on full Golden Path generation success.

The Architectural Reality

GoldenPathSynthesizer.fetchOpenPRs() and renderRecentOpenPrSummary() own the PR snapshot rendering path.
The generated handoff currently includes a Captured at timestamp, but there is no freshness gate that marks the PR section stale when the capture is old.
Prior work removed per-author PR bloat, but the live symptom is now stale PR data remaining in the artifact for days.

The Fix

Add a freshness SLA for Active PR Cycle State / Recent Open PRs inside the generated handoff.
If live PR fetch succeeds, render a current compact PR section capped at 10 with (shown / total).
If live PR fetch fails or the section is older than the SLA, render an explicit stale/degraded marker and do not present old PR data as current.
Add tests that seed an old PR capture and prove regeneration either replaces it with current data or marks PR state stale/degraded.

Contract Ledger

Target Surface	Source of Authority	Proposed Behavior	Fallback / Edge Case	Docs	Evidence
`sandman_handoff.md` Active PR Cycle State	this ticket; `GoldenPathSynthesizer`	PR section is compact, capped, and has current capture metadata.	Live fetch failure renders degraded/stale marker instead of old PR facts.	DreamPipeline update if section contract changes	Unit test with stale seeded artifact and fetch failure.
`renderRecentOpenPrSummary()`	current source	Caps at 10 and renders `(shown / total)` heading.	Empty PR list omits list or states none, without stale prior entries.	JSDoc if changed	Focused renderer test.
PR fetch path	current source	Failure is visible in generated handoff and task outcome.	No silent reuse of old GitHub PR data.	None unless user-facing marker needs docs	Unit test with fetch throwing.

Acceptance Criteria

PR state older than the configured freshness SLA is visibly marked stale/degraded in the generated handoff.
Successful PR fetch replaces stale PR content with current compact data.
Fetch failure does not preserve old PR entries as if current.
Recent Open PRs renders no more than 10 items and displays (shown / total).
Per-author PR sections remain absent.
Tests cover stale seeded handoff, fetch success, and fetch failure.

Out of Scope

REM parser over-cap behavior.
Golden Path strategic summary generation.
Manual markdown cleanup.

#13956 / PR #13957 — prior compact handoff rendering work.
#13962 / PR #13963 — zero-candidate rewrite path.
#13983 — Golden Path strategic handoff contract.

Release classification: boardless urgent Agent OS lifecycle-safety hardening. Retrieval Hint: Active PR Cycle State stale capturedAt Recent Open PRs sandman_handoff

tobiu referenced in commit 6ca9909 - "fix(ai): fail loud on stale handoff PR state (#13985) (#13987)" on Jun 25, 2026, 11:37 AM

tobiu closed this issue on Jun 25, 2026, 11:37 AM